AMD Adopts 28 nm Bulk Manufacturing in 2013

[btarunr]

According to AMD senior VP and CTO Mark Papermaster, the company will adopt the 28 nanometer bulk CMOS silicon fabrication process for its chips in 2013. The bulk process is used to manufacture high-volume and less-complex products, such as motherboard chipset, entry-level APUs, etc. The company already takes advantage of TSMC 28 nm High-Performance process for highly-complex chip designs, such as its Southern Islands GPU family, and will continue using it for its next-generation "Sea Islands" GPUs. In related news, DigiTimes learned through sources that AMD's Sea Islands GPUs have entered tape-out stage, and are on course for a late-2012 volume manufacturing, and early-2013 launch schedule.



Source: DigiTimes

[Assimilator]

Intel's already got working 22nm parts and will have 14nm process perfected by 2013. Why does AMD keep making press releases that only show how far behind it is?

[cadaveca]

Originally Posted by Assimilator

Intel's already got working 22nm parts and will have 14nm process perfected by 2013. Why does AMD keep making press releases that only show how far behind it is?

With a smaller share of the market, it's only natural that they cannot progress as quickly as Intel does, as the dollars for R&D just aren't equal. Expecting similar technology development on an exponentially smaller budget just doesn't seem realistic to me, so AMD being "behind" on prcess tech doesn't bother me one bit.


Why does it bother you?

[Jermelescu]

AMD will have 22/20 nm by late 2014

[hardcore_gamer]

Originally Posted by Assimilator

Intel's already got working 22nm parts and will have 14nm process perfected by 2013. Why does AMD keep making press releases that only show how far behind it is?

And there was a huge performance increase for Intel CPUs when they went from 32nm to 22nm.

[erocker]

Funny, this article isn't even about CPU's.

[sergionography]

Damn the amd graphics department is on a roll!
Process helps but hsa is said to bring much more performance improvements than a process shrink. But let's hope its big enough to compete with Intel process shrink+new architecture. AMD believes it is so lets just hope they dwliver

[suraswami]

I was wondering why the die shrink has to be in a certain order like 65, 45, 32, 28 etc. Instead wasting time on 28 can't they not go directly to like 22 or 2 jumps ahead?

sorry trying to understand why it is like this.

[repman244]

Originally Posted by Assimilator

Intel's already got working 22nm parts and will have 14nm process perfected by 2013. Why does AMD keep making press releases that only show how far behind it is?

AMD doesn't have their own fabs, they rely on GloFo/TSMC.

Originally Posted by hardcore_gamer

And there was a huge performance increase for Intel CPUs when they went from 32nm to 22nm.

There will be when Haswell arrives, IB is just a die shrink of SB + some tweaks. You will never see a lot of improvement with a die shrink as with a proper new architecture.

[theoneandonlymrk]

Originally Posted by btarunr

AMD's Sea Islands GPUs have entered tape-out stage

< Quivers,,, Yes


Now shut the F@%* Up and Take my money.

Originally Posted by suraswami

I was wondering why the die shrink has to be in a certain order like 65, 45, 32, 28 etc. Instead wasting time on 28 can't they not go directly to like 22 or 2 jumps ahead?

sorry trying to understand why it is like this.

its in part due to a set percentage decrease required verses technically possible, and is also based upon what a 300mm wafer or the manufacturers own size diveded up into a certain ammount of chips, and the optical precision possible in the lenses etc of the process(Lythography), a few things go towards a setting of each size, and they are normally at or beyond whats easily possible allready.

[AlB80]

AMD's driver already contains references to Kaveri/Kabini/SeaIsland. It means most of the type-out stage is completed and AMD has some kind of samples.

[trickson]

This is AMD's big press release? Telling every one how far behind they are? Intel has 22nm and is working on 14nm now! WOW AMD you are sooooooooooooo coooooool!

[f22a4bandit]

Originally Posted by erocker

Funny, this article isn't even about CPU's.

Yup. I guess reading is overrated.

[theoneandonlymrk]

Originally Posted by trickson

This is AMD's big press release? Telling every one how far behind they are? Intel has 22nm and is working on 14nm now! WOW AMD you are sooooooooooooo coooooool!

WOW , What a completely new angle i and no one else had thought of

Originally Posted by Assimilator

Intel's already got working 22nm parts and will have 14nm process perfected by 2013. Why does AMD keep making press releases that only show how far behind it is?

are you him??

they are behind not out,and so are intel Gpu yawn ,,ive been down this road , suffice to say main rigs a phenom 960T, and yes intels piss on mine but and what, it folds 24/7 and games well enough

[trickson]

You know what would really impress me ? Is if AMD came out and said we are making a PC that no longer needs a CMOS to start! Now that would impress me. But as for this? LOL COME ON!

[theoneandonlymrk]

Originally Posted by trickson

You know what would really impress me ? Is if AMD came out and said we are making a PC that no longer needs a CMOS to start! Now that would impress me. But as for this? LOL COME ON!

agreed ,, unimpressive

its progress though and thats good..

[acerace]

Originally Posted by trickson

This is AMD's big press release? Telling every one how far behind they are? Intel has 22nm and is working on 14nm now! WOW AMD you are sooooooooooooo coooooool!

Dude, read the article thoroughly.

[Dent1]

Why does anyone here care? As long as the product is fast and value for money I couldnt give a rats ass whether it was 28nm or not. I'm sure the majority of their customer don't give a rats ass either.

[Disruptor4]

It is amazing the stupidity of people who think this is about their CPU's. I guess, as said above, reading really is overrated!

[suraswami]

Originally Posted by acerace

Dude, read the article thoroughly.

Dude it's trickson, just ignore him.

[Dent1]

Originally Posted by Disruptor4

It is amazing the stupidity of people who think this is about their CPU's. I guess, as said above, reading really is overrated!

People have trained themselves to enjoy stories of AMD doing badly.

These people make bad friends, they are the type of people that want to see you do bad too.

[Disruptor4]

Originally Posted by Dent1

People have trained themselves to enjoy stories of AMD doing badly.

These people make bad friends, they are the type of people that want to see you do bad too.

Haha truly intelligent. It's like they think that if they read the header, they know what the story is on about. Even if it was about their CPU's, sure they are behind, but really their performance at the current node is better than what Intel's performance was at that same node.

[seronx]

28-nm GlobalFoundries vs 28-nm TSMC vs 22-nm Intel vs 22-nm AMD-IBM Foundry partnership(before GlobalFoundries was sold)
SRAM Cell Size:
GloFo => ~0.120 um²
TSMC => ~0.127 um²
Intel => ~0.092-0.108 um²
AMD-IBM Fishkill => ~0.1 um²

AMD's foundries aren't that far behind... and since Rambus is working with GlobalFoundries to shrink and speed up SRAM Cells we can expect Steamroller to have faster and bigger L1, L2, L3 caches.

[erocker]

Originally Posted by seronx

AMD's foundries aren't that far behind... and since Rambus is working with GlobalFoundries to shrink and speed up SRAM Cells we can expect Steamroller to have faster and bigger L1, L2, L3 caches.

Not sure if we need larger but faster... Yes please!

[seronx]

Originally Posted by erocker

Not sure if we need larger but faster... Yes please!

L1 needs bigger, L2 and L3 just need more prefetching SRAMs to lower the latency.

Also forgot to mention that 32-nm yields are directly linked to 28-nm yields.

2011 -> 2012 for GlobalFoundries provided a 2x increase in 32-nm PD-SOI yields

2012 -> 2013 yield increase will be whatever is added from that plus the sightly decreased size.
~10% smaller node(vs 32-nm)
~20% smaller SRAM cells(vs 32-nm)
~2x increase from initial 32-nm yields(2011 - Bulldozer to 2013 - Steamroller)
28-nm Bulk CMOS and 28-nm FD-SOI CMOS is also cheaper and tends to have a higher volume/yield rate than 32-nm PD-SOI.